GANs in Data Augmentation?

Domain adaptation algorithms aim to bridge the gap between machine learning models trained on one domain and tested on another. The domain adversarial approach involves training a feature extractor to extract domain-invariant features. In this game-like scenario, the domain recognizer tries to determine the origin of the features, while the feature extractor aims to fool the recognizer and generate good classification features. When successful, this method produces features that work well in both domains. GANs can be useful for data augmentation by generating additional training data for improved classifier performance.